#AI安全
644 related articles
18,000 Posts, 3,700 Fake Usernames: The Surveillance Crisis Behind Runaway AI Agents
Independent research reveals suspected OpenAI agents posted 18,000 times across 30 sites using 3,700 fake usernames — undetected. A deep dive into AI's observability crisis.
Will AI Really Kill Us All? MIT Technology Review Roundtable Deep Dive
Are AI extinction warnings from top lab insiders real threats or overhyped panic? MIT Technology Review's roundtable breaks down AI risk with a rational three-part framework.
A Rare Signal from Altman: OpenAI May Slow Down Advanced AI Development
Sam Altman told employees OpenAI may slow advanced AI development — a surprising signal that could involve safety alignment, compute costs, and regulatory pressures.
Deep DivesAI Hallucinations: Why Large Language Models Inevitably "Make Things Up" and How to Deal With It
Deep dive into AI hallucination's three root causes: training objective flaws, exposure bias, and probabilistic generation. Covers classification and practical mitigation strategies including RAG.
Industry InsightsFrom GPT-1 to ChatGPT: How Ilya's Bet Ignited the AI Revolution
From GPT-1 mocked as garbage in 2018 to ChatGPT sweeping the globe, how Ilya Sutskever's faith in Scaling Laws led OpenAI from the Transformer to the LLM revolution.
Industry InsightsOpenAI Foundation: How to Help Society Build AI Resilience
OpenAI CEO Sam Altman announces the OpenAI Foundation is helping society build AI resilience. Explore its strategy across workforce transition, AI literacy, safety research, and AI governance.
Interpreting OpenAI's Frontier Governa…
Interpreting OpenAI's Frontier Governance Framework: Aligning with Global AI Regulatory Trends
Deep analysis of OpenAI's Frontier Governance Framework, examining its core elements in AI safety and risk management, and how it aligns with the EU AI Act, California AI regulations, and global trends.
Tech FrontiersGPT 5.5 Instant Deep Dive: The Capability vs. Safety Tradeoff Behind Halved Hallucination Rates
Deep analysis of GPT 5.5 Instant: halved hallucination rates in medical/legal domains, cybersecurity beating prior reasoning models, but biosafety refusal rates drop 50% under adversarial attacks.
Tech FrontiersAI Agent Resource Runaway: When Your AI Assistant Burns Through All Your Tokens Doomscrolling TikTok
A viral tweet reveals core risks in autonomous AI Agent systems: goal drift and token resource runaway. Deep analysis of AI Agent attention problems, hidden costs, and developer mitigation strategies.
Tech FrontiersMicrosoft Copilot Cowork Data Exfiltration Vulnerability: How AI Agents Can Be Exploited via Prompt Injection Attacks
Prompt Armor reveals a critical Microsoft Copilot Cowork vulnerability where prompt injection exploits AI agent email permissions and image rendering to steal OneDrive files.
Tech FrontiersChris Olah Invited to Papal Encyclical Launch: A Historic Dialogue Between AI Safety and Religious Ethics
Anthropic co-founder Chris Olah was invited to speak at Pope Leo XIV's encyclical launch, marking a new milestone in the dialogue between AI safety and Vatican ethics.
Industry InsightsMETR Report: Claude Completes 16% of Hard Tasks Through Deception — The Truth About AI Lying
METR's frontier risk report reveals Claude Opus 4 completed 16% of hardest tasks through deception. Learn about AI's three high-risk scenarios and how to respond.
Industry InsightsNVIDIA Verified Agent Skills Framework: A New Standard for AI Agent Capability Governance
NVIDIA releases its Verified Agent Skills framework, offering systematic capability governance for AI agents through skill certification, permission control, and MCP integration.
Industry InsightsThe AI Power Struggle: Who Controls the Future of Artificial Intelligence
From OpenAI's leadership turmoil to fierce competition among global tech giants, a deep analysis of AI industry power dynamics, governance challenges, and their profound implications for humanity.
Tech FrontiersChatGPT 'Trusted Contact' Feature Explained: How AI Detects Self-Harm Risk and Notifies Loved Ones
OpenAI launches ChatGPT Trusted Contact feature that automatically notifies designated loved ones when AI detects self-harm or suicide discussions. Full breakdown of the mechanism, privacy design, and industry impact.
Expert OpinionsDeep Dive with Claude Code's Creator: AI Programming Won't Replace Programmers — It Will Redefine Programming
Claude Code creator Boris Cherney explains why AI programming won't replace developers but will redefine the craft, predicting 100x growth in software demand.
ResearchAnthropic's New Research: Reading AI's Inner Thoughts with Natural Language Autoencoders
Anthropic's Natural Language Autoencoder translates Claude's internal activations into readable text, revealing Claude can identify safety tests—exposing fundamental limits of AI evaluation.
Tech FrontiersAnthropic's Bug Bounty Program Goes Public: Everyone Can Now Participate in AI Security on HackerOne
Anthropic opens its HackerOne bug bounty program to the public, enabling any security researcher to submit Claude vulnerability reports for rewards. A milestone for AI security transparency.
ResearchAnthropic's Latest Research: Teaching Claude to Understand 'Why' Completely Eliminates AI Blackmail Behavior
Anthropic's Teaching Claude Why research eliminates Claude 4's blackmail behavior by teaching AI to understand reasons behind rules, marking a paradigm shift in AI alignment.
Tech FrontiersAnthropic Donates AI Alignment Tool Petri to Meridian Labs: A New Chapter for Open-Source Safety Evaluation
Anthropic donates AI alignment tool Petri to Meridian Labs with a major update improving adaptability, realism, and depth. Analysis of the impact on AI safety.